Problem
Manual QA, rising AWS spend, and throughput constraints slowed lead flow.
Approach
- Event-driven serverless architecture on AWS (Lambda, API Gateway, Step Functions, SQS, DynamoDB/Aurora, S3).
- IaC with CDK/Pulumi and CI/CD (GitHub Actions); observability via CloudWatch + Sentry.
- LLM data validation & text→JSON extraction with guardrails and schema checks.
- Real-time BI with QuickSight; nightly model retraining (PyTorch/scikit-learn).
Results (Quantified)
- Scaled outreach from ~100k → 2M prospects/month; +112% appointments.
- 40% AWS cost reduction via right-sizing & architecture changes.
- Faster lead flow; automated QA replaced manual bottlenecks.
Architecture (Thumbnail)
View full-size SVG
(Sanitized, representative architecture.)
Tech Stack
AWS: Lambda, API Gateway, Step Functions, SQS/SNS, DynamoDB, Aurora/RDS, S3, CloudWatch
Data/ML: Postgres, ETL/ELT, QuickSight, PyTorch, scikit-learn
Dev: Node.js/TypeScript, Python (FastAPI and for ML model creation / training), Docker, CI/CD (GitHub Actions)
AI/LLM: OpenAI/ChatGPT, guardrails, function calling, schema validation